portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2024-11-10 07:55:35 +00:00
Code Releases Activity

Merge branch 'master' of https://github.com/paboyle/Grid

Browse Source
This commit is contained in:
Peter Boyle 2015-07-01 22:51:04 +01:00
commit 31a0c8d783
6 changed files with 29 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
benchmarks/Benchmark_memory_bandwidth.cc
View File

@ -20,7 +20,7 @@ int main (int argc, char ** argv)
std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl; std::cout << "Grid is setup to use "<<threads<<" threads"<<std::endl;
std::cout << "===================================================================================================="<<std::endl; std::cout << "===================================================================================================="<<std::endl;
std::cout << "= Benchmarking fused AXPY bandwidth"<<std::endl; std::cout << "= Benchmarking fused AXPY bandwidth ; sizeof(Real) "<<sizeof(Real)<<std::endl;
std::cout << "===================================================================================================="<<std::endl; std::cout << "===================================================================================================="<<std::endl;
std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl; std::cout << " L "<<"\t\t"<<"bytes"<<"\t\t\t"<<"GB/s"<<"\t\t"<<"Gflop/s"<<std::endl;
std::cout << "----------------------------------------------------------"<<std::endl; std::cout << "----------------------------------------------------------"<<std::endl;

2
lib/algorithms/approx/Remez.h
View File

@ -15,7 +15,7 @@
#ifndef INCLUDED_ALG_REMEZ_H #ifndef INCLUDED_ALG_REMEZ_H
#define INCLUDED_ALG_REMEZ_H #define INCLUDED_ALG_REMEZ_H
#include <algorithms/approx/bigfloat.h> #include <algorithms/approx/bigfloat_double.h>
#define JMAX 10000 //Maximum number of iterations of Newton's approximation #define JMAX 10000 //Maximum number of iterations of Newton's approximation
#define SUM_MAX 10 // Maximum number of terms in exponential #define SUM_MAX 10 // Maximum number of terms in exponential

16
scripts/configure-commands
View File

@ -16,10 +16,16 @@ echo -e $RED ==============================
echo -e $YELLOW echo -e $YELLOW
case $WD in case $WD in
g++-sse4) g++-avx)
CXX=g++ ../../configure --enable-simd=AVX CXXFLAGS="-mavx -O3 -std=c++11" LIBS="-lgmp -lmpfr" --enable-comms=none
;;
g++-avx-openmp)
CXX=g++ ../../configure --enable-simd=AVX CXXFLAGS="-mavx -fopenmp -O3 -std=c++11" LIBS="-fopenmp -lgmp -lmpfr" --enable-comms=none
;;
g++5-sse4)
CXX=g++-5 ../../configure --enable-simd=SSE4 CXXFLAGS="-msse4 -O3 -std=c++11" LIBS="-lgmp -lmpfr" --enable-comms=none CXX=g++-5 ../../configure --enable-simd=SSE4 CXXFLAGS="-msse4 -O3 -std=c++11" LIBS="-lgmp -lmpfr" --enable-comms=none
;; ;;
g++-avx) g++5-avx)
CXX=g++-5 ../../configure --enable-simd=AVX CXXFLAGS="-mavx -O3 -std=c++11" LIBS="-lgmp -lmpfr" --enable-comms=none CXX=g++-5 ../../configure --enable-simd=AVX CXXFLAGS="-mavx -O3 -std=c++11" LIBS="-lgmp -lmpfr" --enable-comms=none
;; ;;
icpc-avx) icpc-avx)
@ -52,6 +58,12 @@ CXX=clang++ ../../configure --enable-simd=AVX2 CXXFLAGS="-mavx2 -mfma -O3 -std=c
clang-avx-openmp) clang-avx-openmp)
CXX=clang-omp++ ../../configure --enable-simd=AVX CXXFLAGS="-mavx -fopenmp -O3 -std=c++11" LDFLAGS="-fopenmp" LIBS="-lgmp -lmpfr" --enable-comms=none CXX=clang-omp++ ../../configure --enable-simd=AVX CXXFLAGS="-mavx -fopenmp -O3 -std=c++11" LDFLAGS="-fopenmp" LIBS="-lgmp -lmpfr" --enable-comms=none
;; ;;
clang-xc30)
CXX=$HOME/Clang/install/bin/clang++ ../../configure --enable-simd=AVX CXXFLAGS="-mavx -O3 -std=c++11 -I/opt/gcc/4.9.2/snos/include/g++/x86_64-suse-linux/ -I/opt/gcc/4.9.2/snos/include/g++/ " LDFLAGS="" LIBS="-lgmp -lmpfr" --enable-comms=none
;;
clang-xc30-openmp)
CXX=$HOME/Clang/install/bin/clang++ ../../configure --enable-simd=AVX CXXFLAGS="-mavx -fopenmp -O3 -std=c++11 -I/opt/gcc/4.9.2/snos/include/g++/x86_64-suse-linux/ -I/opt/gcc/4.9.2/snos/include/g++/ " LDFLAGS="-fopenmp" LIBS="-lgmp -lmpfr" --enable-comms=none
;;
clang-avx2-openmp) clang-avx2-openmp)
CXX=clang-omp++ ../../configure --enable-simd=AVX2 CXXFLAGS="-mavx2 -mfma -fopenmp -O3 -std=c++11" LDFLAGS="-fopenmp" LIBS="-lgmp -lmpfr" --enable-comms=none CXX=clang-omp++ ../../configure --enable-simd=AVX2 CXXFLAGS="-mavx2 -mfma -fopenmp -O3 -std=c++11" LDFLAGS="-fopenmp" LIBS="-lgmp -lmpfr" --enable-comms=none
;; ;;

10
scripts/configure-cray Executable file
View File

@ -0,0 +1,10 @@
#!/bin/bash
DIRS="g++-avx-openmp g++-avx clang-xc30 clang-xc30-openmp"
for D in $DIRS
do
mkdir -p builds/$D
cd builds/$D
../../scripts/configure-commands $D
cd ../..
done

2
scripts/cray-modules Normal file
View File

@ -0,0 +1,2 @@
module swap PrgEnv-cray PrgEnv-intel
module swap intel/14.0.4.211 intel/15.0.2.164

2
tests/Test_dwf_hdcr.cc
View File

@ -1,6 +1,6 @@
#include <Grid.h> #include <Grid.h>
#include <algorithms/iterative/PrecGeneralisedConjugateResidual.h> #include <algorithms/iterative/PrecGeneralisedConjugateResidual.h>
#include <algorithms/iterative/PrecConjugateResidual.h> //#include <algorithms/iterative/PrecConjugateResidual.h>
using namespace std; using namespace std;
using namespace Grid; using namespace Grid;